2D(PC)2A Based Dimensionality Reduction of Textural Feature for Face Recognition with a Single Training Sample

Face recognition system works badly in practical applications because only single training sample image per person is stored in the system owing to hard collecting training samples. We present a novel face recognition scheme with single training sample using 2D Gabor filter and 2D(PC)2A under varying light conditions. Firstly, 2D texture feature extract with Gabor filter captures the properties of spatial localization, orientation selectivity, and spatial frequency selectivity to cope with the variations in illumination. Secondly, 2D(PC)2A is to extract statistical texture features under one training sample. Finally matrix-based similarity nearest neighbor classifier is used to classify a new face for recognition. Some experiments are implemented to testify the feasibility of the proposed scheme.